Registered Switch 2 input-to-broker matrix¶

This is software composition evidence, not physical, Windows API, game or latency acceptance. The old ViiperCompatibilityMatrixTests selected numerical seeds using a physical-family enum and built packets from hand-filled DS4State. Its names now say that explicitly. It never proved physical-family integration.

Switch2ProductionInputMatrixTests reuses the existing registered gyro-mapping fixture. The path is Common05 bytes -> real decoder/calibration/profile projection -> runtime device with a production registration transaction and slot host -> SetCurveAndDeadzone/MapCustom -> final broker packet encoder. Registration supplies the logical report handler; tests do not install another mapper. Fake OS discovery/transport pumps and profile persistence prevent any controller, keyboard, mouse, installed configuration or driver access.

Coverage¶

Seven source modes cross six target types:

• 42 routes check applicable ordinary buttons, both digital triggers at full scale, repeated held state, releases, represses, exact field masks and terminal neutral before profile teardown. Capture has no native representation on Xbox 360/Sony here; Xbox One Share and Switch Capture are asserted explicitly.
• 42 routes check each raw stick axis at 0, 1, 2047, 2048, 2049, 4094, 4095. Independent expected arithmetic verifies center/endpoints, near-center detail, unused-half neutrality, vertical/horizontal rotation, Xbox signed-16 and Switch unsigned-12 precision, and Sony's final signed-byte quantization.
• 21 explicit extra-control binding cases each check all six outputs: Pro GL/GR, Capture, C, horizontal mini-controller paddle roles and four joined SL/SR identities. Mapping to A must not also leak the old native button; release and terminal removal must release the binding.
• Five Joy-Con-mode cases reject manufacture of controls from Pro-only rear bits. The existing raw/unknown metadata policy is not broadened by SDL's inclusion of those bits in a combined handler.

The bit/orientation facts are rechecked against the existing pinned references: upstream SDL c71abd08605b8bb7078372307a93274725c99fe0, src/joystick/hidapi/SDL_hidapi_switch2.c combined/mini handlers; and Switch2Connect 61ac6642ce12fe7217e38a860b14863b18ca7e28, src/controller.py btn_states (GL/GR gated by is_pro; physical rails separate). See Joy-Con protocol/provenance. No reference repository was modified and no PadForge code was copied. Golden output fields are the existing DS4Windows/VIIPER broker contracts, not raw USB/GIP reports; Xbox One's versioned contract is asserted with version, size and reserved tail checks as well.

Limits¶

Transport pumps, persisted-profile loading and broker sockets are substitutes or absent. New decoder sessions seed the fixtures; this matrix does not prove counter continuity, arrival ordering, skew arbitration, BLE initialization, association or physical disconnect. It does not exercise Windows virtual-device selection/creation, scheduler admission, USB/IP terminal completion, game polling or haptic output. Existing lifecycle/interop tests cover separate boundaries.

The axis cases use fallback calibration and the fixture's default/reset stick configuration. Real calibration, all custom transforms, all source firmware, live orientation edits, optical/gyro modes, unknown layouts, legacy physical families, every source-target feedback route, hardware and game acceptance still need their own evidence. Passing this matrix is not full controller support.

This tranche changes tests and documentation only. Live b56 and staged b58 payloads are unchanged. Exact TRX results are recorded in VIIPER's dated ledger.